US20070245044A1 - System of interconnections for external functional blocks on a chip provided with a single configurable communication protocol - Google Patents

Abstract

System of interconnections for external functional blocks on a chip provided with a single configurable communication protocol The system of interconnections (20) for external functional blocks on a chip provided with a single configurable communication protocol, comprises two physically separate communication networks (21, 22): a request network (21) for transmitting request messages from an initiating block (23, 24, 25, 26) to a recipient block (27, 28, 29, 30, 31) and a response network (22) for transmitting response messages from a recipient block (27, 28, 29, 30, 31) to an initiating block (23, 25, 26). The response messages include additional information making said request (21) and response (22) networks able to respectively manage the request messages and the response messages independently.

Description

• The present invention relates to a system of interconnections for external functional blocks on a chip provided with a single configurable communication protocol.
• The ongoing trend in technology means that increasingly complex systems made up of numerous different functional blocks can be integrated. These functional blocks are developed by different teams using different methods and communication protocols, these functional blocks being called intellectual properties.
• It is therefore essential for these functional blocks to be able to intercommunicate, despite their different operating characteristics, such as the size of the data used and the frequencies used.
• Furthermore, since the communicating elements are becoming more and more numerous, the systems need to be flexible or adaptable enough to support this trend.
• Moreover, the cost of the wires in a system of interconnections has become dominant over the cost of the logic gates. Also, assessing an interconnection network architecture on silicon takes into account the conditions of use of the wires and their cost effectiveness (quantity of information transported in relation to the cost). Furthermore, since the time to market for new products is tending to shorten, the integration of the functional blocks needs to be more and more rapid. Also, the system or network of interconnections must be easy to implement.
• Solutions exist for interconnecting IP functional blocks in a system of interconnections on a semiconductor chip, and are mostly based on data buses. These networks lack the necessary flexibility for rapid adaptation to technological changes and for increasing the number of communicating elements.
• Such network architectures do not easily allow adaptation to the changing application-oriented requirements.
• Thus, one object of the invention is to propose a system of interconnections for external functional blocks on a chip provided with a single configurable communication protocol, that can easily be adapted according to the changing application-oriented requirements.
• Thus, according to one aspect of the invention, there is proposed a system of interconnections for external functional blocks on a chip provided with a single configurable communication protocol. The system comprises two physically separate communication networks, a request network for transmitting request messages from an initiating block to a recipient block and a response network for transmitting response messages from a recipient block to an initiating block. Furthermore, the response messages include additional information making said request and response networks able to respectively manage the request messages independently of the response messages.
• Since the request and response networks are capable of respectively managing the request messages independently of the response messages, they are autonomous. A response packet does not need to know the associated request, a change of size or frequency of the response packet is therefore possible without additional information and without saving the context between requests and responses within the system or network of interconnections.
• According to an embodiment, said protocol is organized in a plurality of independent layers. Said independent layers comprise a transport layer defining the structure of a data packet, and a physical layer defining the signals interchanged at the interfaces of the elements of the system of interconnections. Address, data size and operation information is contained in the headers of the data packets.
• Since the protocol is organized in separate layers, the rules that define the behaviour of a layer are independent of the way in which the preceding layer is implemented. This means it is easier to make optimizations on each layer, independently of the choices that might have been made on the other layers.
• Furthermore, this makes it possible to reduce the number of wires needed, by using the same wires to convey both the data of the packets, and data representative of address and control information.
• According to an embodiment, said additional information of the response message data packets includes information representative of the operation carried out, and information representative of the size of the data associated with said operation carried out.
• For example, the operation is a data read or a data write.
• According to an embodiment, following a read of data that is incomplete or unaligned in relation to the memory space accessed in said read, the packets of the response message associated with said read include, at a predetermined position, said data that has been read.
• According to an embodiment, following a read of data that is incomplete or unaligned in relation to the memory space accessed in said read, the packets of the response message associated with said read include information for aligning said data that has been read in relation to said memory space accessed.
• According to an embodiment, the interface between two elements of the system of interconnections comprises:
• • means for receiving or transferring a clock timing signal,
  • means for receiving or transferring signals for stopping or starting said elements independently,
  • means of transferring flow control signals relating to the clock cycles,
  • means of transferring data packet delimiting signals, and
  • means of transferring signals of configurable size for transmitting the content of the data packets.
• According to an embodiment of the invention, the interface between two elements of the system of interconnections also comprises signal generation means for favouring the transmission of some messages over others.
• According to an embodiment, the headers of the data packets include an available space reserved for the use of the initiating blocks.
• According to an embodiment, the data packets are of variable length, equal to a multiple of an elementary quantity of data. A system link can be programmed by the ratio between the quantity of data that it can transmit in each clock cycle and said elementary quantity of data.
• According to an embodiment, the links include means of changing the clock frequency, and/or means of deleting data holes in the data packets in case of predictable hole ratios, and/or means of changing the serialization of the packets.
• According to an embodiment, the elements of the system of interconnections are configurable by software means via a separate network.
• For example, said separate network is able to operate at a low bit rate compared to the average bit rate in the system of interconnections.
• Another aspect of the invention proposes generators of configurable IP blocks for constructing a system in a modular fashion as described previously, using standard interfaces.
• According to an embodiment, the generators comprise:
• • a generator of configurable protocol IP blocks,
  • generators of network interface IP blocks between said configurable protocol and other communication protocols,
  • generators of switch IP blocks,
  • generators of IP blocks for changing clock, domain, compacting data packets, transporting packets over long distances, and programming ratios between the quantity of data that a link can transmit in each clock cycle and said elementary quantity of data, and
  • generators of IP blocks of a separate network for configuring the elements of the system of interconnections.
• Other aims, features and advantages of the invention will become apparent from reading the description that follows, of a few by no means limiting examples, and with reference to the appended drawings, in which:
• FIG. 1 illustrates the configurable protocol organized in a plurality of independent layers, according to one aspect of the invention; and
• FIG. 2 illustrates a system of interconnections for external functional blocks on a chip, according to one aspect of the invention.
• As diagrammatically illustrated inFIG. 1, the protocol is defined on a number of independent layers, in this case on three layers: the physical layer, the transport layer and the transactional layer.
• The transport layer is used to control the interchange of data packets. A data packet is made up of an integer number of cells of equal size.
• A cell is an elementary quantity of data, the size of which is a parameter of the communication protocol.
• The size of the packet header is one or two cells, depending on whether it is a response packet or a request packet. For example, the header of a request data packet contains the address of the recipient element in the network and the storage address in the addressing space of the recipient element, which does not need to be present in the header of the associated response data packet. The addressing space of the recipient element corresponds to a range of addresses that can be accessed, or at which a hardware device can be reached. These addresses can be physical or virtual.
• A data packet header always contains the address of the initiating element in the system or network of interconnections, and an information tag, used to identify the various data interchanges performed by one and the same initiating element.
• In practice, when an initiating element can communicate with several recipient elements, it needs to be able to identify the origin of the response data packets that it receives and it cannot therefore control the sequencing of the packets if they come from different recipient elements via different paths.
• Placing the address of the recipient element in the header of a response packet can be costly if there is a large number of recipient elements.
• However, since, at a given instant, the number of recipient elements corresponding with the initiating element is limited, each of these recipient elements is assigned an identifier, or a tag, on a small number of bits.
• This tag is associated with the request data packet, and when the recipient element sends a corresponding response data packet, it attaches the corresponding tag to it.
• The transport layer is separated from the physical layer by placing in the header of a packet all the address and control information that concerns the packet, but not information concerning the manner in which it will physically be transported.
• The header of a packet contains an operation code specifying the type of operation carried out or to be carried out (for example, read type or write type), the address of the initiating element in the network and a tag identifying the corresponding recipient element.
• The header of a data packet can also contain, depending on the case, the address of the recipient element, the address corresponding to the operation carried out in the addressing domain of the recipient element, the length of the data, the transport control bits (corresponding, for example, to an exclusive or secured transaction), and any other information encoded on available bits reserved for the use of the initiating elements.
• The communication protocol defines, for a data packet, the positioning of the various fields in the header of the data packet according to the operation codes.
• The separation of the request data packets and the response packets, with autonomous response packets, is performed by placing in the response data packet all the information needed to avoid saving contexts, in particular on changes of format on the data packets, in the elements that make up the network of interconnections.
• One way of ensuring the autonomy or independence of the response data packets with respect to the request data packets is to have an operation code defining the transaction type (for example, read type or write type) and the associated quantity of data in the case of a read.
• Furthermore, on a read, it may be that information that is incomplete or unaligned in relation to the memory space accessed at the time of the read in the recipient element is accessed.
• In practice, the data is normally read in the format of the recipient element, and transmitted as such to the corresponding initiating element. Now, if the request requires a particular positioning of the data for the corresponding response, this positioning information must be determined from an associated response data packet.
• It is, for example, possible to position the data that is read in a response data packet, at a predetermined position, and it is the initiating element that realigns the data on arrival by knowing the predetermined position.
• It is also possible to insert into the header of a response data packet information for aligning the data that is read relative to the memory space accessed.
• Any physical interface is independent of the message type, because, for the physical transport, all the cells or elementary data quantities are equivalent. The physical interface mainly defines the nature of the signals in the links between the elements that make up the network of interconnections.
• The physical interface comprises the signals essential to the physical transport of the data during a clock cycle (FIG. 1), such as:
• • a clock timing signal;
  • independent stop and restart signals for an element of the system of interconnections including at least one reset signal. These signals can include a power supply control signal and/or a signal for initializing flow control on the link;
  • flow control signals relating to the clock cycles, for deleting the data packet storage queues at the nodes of the network, or for having holes in the data packets. These signals can also include a signal from the sender to the receiver of a data packet, representing data validity information, and a signal from the receiver to the sender of a data packet, comprising information representing the ability of the receiver to receive the data. A receiver that is not ready to receive data can block the sending of data by the sender;
  • signals delimiting data packets;
  • signals of configurable size for transferring data during a clock cycle, not necessarily corresponding to the size of a cell or elementary data quantity. The number of cells conveyed in each cycle, or word, defines the degree of these implementations of the data packet. The degrees of these implementations that are most practical are a quarter, a half, one, two or four cells conveyed in each clock cycle. The cell size depends on the parameters of the communication protocol;
  • any signals linked to the service quality of a link or, in other words, signals for favouring the transmission of some messages over others, for favouring the passage of data packets considered priority.
• When necessary, the control information present in the header of a request data packet can be analysed by the recipient element and retransmitted, modified or not, to the initiating element in the associated response data packet.
• The headers of the data packets include an available space reserved for the use of the initiating blocks.
• For example, if a number of transactions of an initiating element are aggregated in a single request data packet at the network interface, instead of saving how the requests have been aggregated in a context queue, the information is encoded on the private bits of the reserved available space in the header of the request data packet.
• When the corresponding response data packet is received by the initiating element, the private data bits that have been transmitted to it are used to separate the response and so terminate this set of transactions initiated by the initiating element.
• Normally, the cell size is fixed by the constraints on the data sizes rather than the constraints on the sizes of the data packet headers. The header therefore normally includes a few unused bits, which can thus be exploited to advantageously replace context memories in the interfaces of the system of interconnections.
• The content of the packet does not depend on the manner in which it is transported. The request and response data packets do not need to have the same serializations, which are defined for each link in turn, by choosing the physical width of the link.
• On changes of format, the widths of the links and the clock timing frequency can change, but the bit rates are adjusted to the slowest elements by the flow control.
• For a change of frequency, a FIFO type queue is used, with two clocks, the size of which is sufficient never to needlessly lose cycles corresponding to the round-trip time of the flow control in the local loop, counted as a number of cycles of the slowest clock. A size of a few words is sufficient for such a queue.
• When the frequency or the size of the link is increased, data holes appear within the data packets, which can reduce the efficiency of certain recipient elements, such as a DRAM memory controller.
• These data holes can be eliminated using a FIFO type queue, which stores all of a data packet before transmitting it.
• However, this is costly in terms of implementation area, and in terms of latency, because, in this case, the queue needs to have a size equal to the maximum size of a data packet supported by the communication protocol.
• However, when the ratio of data holes in the data packet can be predicted, the size of the compaction queue can be reduced using the size information present in the header of the data packets.
• The use of this information and the knowledge of the ratio of data holes makes it possible to anticipate the sending of the data packet with no data hole, when this is possible, and without waiting for the end of the data packet.
• For example, if the bit rate is doubled, the size of the queue and the latency can be reduced by a factor of two.
• In a system of interconnections made up of point-to-point links, the paths between the initiating elements and the recipient elements are partially shared in order to make best use of the wires.
• There are therefore data flow convergence and redistribution points at the nodes of the network of interconnections, which are implemented by means of switches.
• When these nodes form routing cycles, there is a risk of deadlock in the network of interconnections by the appearance of a string of data packets that are mutually blocked. In such a deadlock situation, it is then necessary to reset all of the system.
• It is possible to guarantee, by construction, the elimination of this risk of deadlock, because, with the separation of the requests and the responses, to avoid such loops, it is enough to avoid the presence of such loops in the topology of each of the request and response networks.
• As illustrated inFIG. 2, the system ofinterconnections20 comprises arequest network21 and aresponse network22 that are physically separate.
• These two physically separate networks do not include common elements, and are represented separately for greater clarity, even if, in reality, these two networks intersect.
• Message initiating elements23,24,25 and26 can transmit request messages torecipient elements27,28,29,30 and31 via therequest network21.
• Therecipient elements27,28,29,30 and31 can transmit response messages to the initiatingelements23,25,26, which require a response on their-part, via the response networks22.
• For greater clarity, the twonetworks21 and22 have been represented side by side, and themessage initiating elements23,24 and26 requiring a response to a request message sent to arecipient element27,28,29,30 and31 have been shown duplicated for greater clarity.
• Therequest network21 has twoswitches32,33 and ademultiplexer34 and fourformat converters35,36,37 and38.
• Therequest network21 also comprises network interfaces39,40,41 and42, respectively dedicated to the initiatingelements23,24,25 and26.
• Furthermore, therequest network21 comprises network interfaces43,44,45,46 and47, respectively dedicated to therecipient elements27,28,29,30 and31.
• Theresponse network22 comprises fourswitches48,49,50 and51, onemultiplexer52, and threeformat converters53,54 and55. Furthermore, theresponse network22 comprises network interfaces56,57,58,59 and60, respectively dedicated to therecipient elements27,28,29,30 and31, and network interfaces61,62 and63, respectively dedicated to the initiatingelements23,25 and26, which require a response message after a request message has been sent.
• Therequest network21 comprises threesubnetworks64,65 and66, each having their own clock domain.
• Theresponse network22 comprises threesubnetworks67,68 and69, each having their own clock domain.
• Thesubnetwork64 comprises the network interfaces39,40,43,44 and45, theswitch32 and theformat converters35 and36.
• The network interfaces39 and40, and theformat converter35, can respectively transmit data packets, to theswitch32, via thelinks70,71 and72.
• Theswitch32 can transmit data packets to the network interfaces43,44 and45, and to the format converter36, respectively vialinks73,74,75 and76. Furthermore, theswitch32 can transmit data packets to another element via alink77.
• Thesubnetwork65 comprises the network interfaces41,42 and46, thedemultiplexer34, and theswitch33. The network interface41 can transmit data packets to thedemultiplexer34 via a link78.
• Thedemultiplexer34 can transmit data packets to theformat converters35 and37, and to theswitch33, respectively vialinks79,80 and81. The network interface42 can transmit data packets to theswitch33 via alink82, and the format converter36 via alink82a.
• Theswitch33 can transmit data packets to thenetwork interface46 and to theformat converter38, respectively vialinks83 and84.
• Thesubnetwork66 comprises theformat converters37 and38, and thenetwork interface47. Theformat converters37 and38 can respectively transmit data packets to thenetwork interface47 vialinks85 and86.
• Thesubnetwork67 comprises the network interfaces56,57 and58, theswitches48,49 and50, theformat converters53 and54 and thenetwork interface61.
• The network interfaces57 and58 can transmit data packets to theswitch49, respectively vialinks87 and88. Thenetwork interface56 can transmit data packets to theswitch48 via alink89.
• Theswitch48 can also receive data packets from an element external to theresponse network22 via alink90.
• Theswitches48 and49, and theformat converter53 can respectively transmit data packets to theswitch50 vialinks91,92 and93. Theswitch50 can respectively transmit data packets to thenetwork interface61 and to the format converter54 vialinks94 and95.
• Thesubnetwork68 comprises the network interfaces59,62 and63, the switch-51 and themultiplexer52.
• The network interface59 and the format converter55 can respectively transmit data packets to theswitch51 vialinks96 and97.
• Theswitch51 can respectively transmit data packets to theformat converter53, to themultiplexer52 and to the network interface63 vialinks98,99 and100.
• The format converter54 can transmit data packets to themultiplexer52 via alink101.
• Furthermore, themultiplexer52 can transmit data packets to the network interface62 via a link102.
• Thesubnetwork69 comprises thenetwork interface60 and the format converter55, and a link103 enabling thenetwork interface60 to transmit data packets to the format converter55.
• Serialization s denotes the ratio of the width of the link (or word size) used to the size of the cell. For thesubnetworks64 and67, the serialization s is one, for thesubnetworks65 and66, the serialization s is two, and for thesubnetworks66 and69, the serialization s is four.
• InFIG. 2, the paths used for a transaction can be followed from left to right.
• The initiating element (on the left) sends a request data packet which passes through the request network along a determined path to the recipient element.
• The latter normally returns a response data packet through the response network for the same initiating element (duplicated on the right of the figure).
• However, some initiating elements, such as memories with direct access, or DMA, do not require a response from a recipient element, and are not duplicated on the right ofFIG. 2, such as the initiatingelement24.
• The request and response networks are not symmetrical. Also, it is advantageous to be able to make independent optimizations of the request and response networks.
• Some elements that make up the network or system of interconnections can advantageously be configured by means of application software, by being able to program the values of software registers accessed by means of a network dedicated to this purpose.
• Such programming is mainly done on system initialization and supports very slow bit rates. Also, it is interesting to use, to this end, a very low cost network. Each software register has its address in this dedicated network, which can be accessed by the request and response networks, with the presence, if necessary, of an interface between this dedicated network and the request and response networks.
• Also, the separation of the routing in the network allows a certain independence in the optimization of each link, which facilitates the design and implementation of a set of IP blocks, or intellectual property blocks, for the elements that make up the network of interconnections.
• These IP blocks are configurable and can be assembled by simply joining up standard physical interfaces end to end to form a transport network. So, this “library” will contain at least the following elements:
• • a utility for programming the communication protocol;
  • a generator of IP blocks that can be configured for each type of network interface supported;
  • a generator of IP blocks for configurable switches, including as parameters the number of inputs, the number of outputs, the types of routing tables, the layers of a pipeline architecture and the arbitration types;
  • transport IP blocks for changing the frequency and the format of the links, for compacting the packets or for transporting the packets over long distances with or without pipeline; and
  • means required to access the configuration registers via the low cost programming network, also called service network, to configure the IP blocks generated.
• Each generator can, for example, produce descriptions in Verilog or VHDL language that can be synthesized by the standard CAD tools.
• These generators are coded in a language that enables the parameter space to be described easily in a structured manner.
• For example, a generator can support lists of parameters, which can themselves be configured as in the case of the generator of switch IP blocks, for which it is possible to choose an arbitration type for each output, each type of arbiter having its own parameters.

Claims (15)

1. System of interconnections for external functional blocks on a chip provided with a single configurable communication protocol, the system comprising two physically separate communication networks, a request network for transmitting request messages from an initiating block to a recipient block and a response network for transmitting response messages from a recipient block to an initiating block, wherein the response messages include additional information making said request and response networks able to respectively manage the request messages and the response messages independently.

2. System according toclaim 1, wherein said protocol is organized in a plurality of independent layers, said independent layers comprising a transport layer defining the structure of a data packet, and a physical layer defining the signals interchanged at the interfaces of the elements of the system of interconnections, address, data size and operation information being contained in the headers of the data packets.

3. System according toclaim 1, wherein said additional information of the response message data packets includes information representative of the operation carried out, and information representative of the size of the data associated with said operation carried out.

4. System according toclaim 1, wherein the operation is a data read or data write.

5. System according toclaim 1, wherein, following a read of data that is incomplete or unaligned in relation to the memory space accessed in said read, the packets of the response message associated with said read include, at a predetermined position, said data that has been read.

6. System according toclaim 1, wherein, following a read of data that is incomplete or unaligned in relation to the memory space accessed in said read, the headers of the packets of the response message associated with said read include information for aligning said data that has been read in relation to said memory space accessed.

7. System according toclaim 1, wherein the interface between two elements of the system of interconnections comprises:

means for receiving or transferring a clock timing signal;

means for transferring or receiving signals for stopping or starting said elements independently;

means of transferring flow control signals relating to the clock cycles,

means of transferring data packet delimiting signals, and

means of transferring signals of configurable size for transmitting the content of the data packets.

8. System according toclaim 7, in which the interface between two elements of the system of interconnections also comprises signal generation means for favoring the transmission of some messages over others.

9. System according toclaim 1, wherein the headers of the data packets include an available space reserved for the use of the initiating blocks.

10. System according toclaim 1, wherein the data packets are of variable length, equal to a multiple of an elementary quantity of data, a system link being programmable by the ratio between the quantity of data that it can transmit in each clock cycle and said elementary quantity of data.

11. System according toclaim 1, wherein the links include means of changing the clock frequency, and/or means of deleting data holes in the data packets in case of predictable hole ratios, and/or means of changing the serialization of the packets.

12. System according toclaim 1, wherein the elements of the system of interconnections are configurable by software means via a separate network.

13. System according toclaim 12, in which said separate network is able to operate at a low bit rate compared to the average bit rate in the system of interconnections.

14. Generators of configurable IP blocks for constructing a system in modular fashion according toclaim 1, using standard interfaces.

15. Generators according toclaim 14, comprising:

a generator of configurable protocol IP blocks;

generators of network interface IP blocks between said configurable protocol and other communication protocols, generators of switch IP blocks,

generators of IP blocks for changing clock domain, compacting data packets, transporting packets over long distances, and programming ratios between the quantity of data that a link can transmit in each clock cycle and said elementary quantity of data, and

generators of IP blocks of a separate network for configuring the elements of the system of interconnections.

Images